From 1dae7ffa0b11293fa3294fc25cd38055915fcf19 Mon Sep 17 00:00:00 2001
From: James Darnley <jdarnley@obe.tv>
Date: Mon, 28 Nov 2016 18:22:17 +0100
Subject: [PATCH] avcodec/h264: mmx 4:2:2 idct add8 function

2.87 times faster (1830 vs. 638 cycles)
---
 libavcodec/x86/h264_idct.asm  | 32 ++++++++++++++++++++++++++++++++
 libavcodec/x86/h264dsp_init.c |  7 ++++++-
 2 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/libavcodec/x86/h264_idct.asm b/libavcodec/x86/h264_idct.asm
index 27222cbc3c..c36fea59b0 100644
--- a/libavcodec/x86/h264_idct.asm
+++ b/libavcodec/x86/h264_idct.asm
@@ -697,6 +697,38 @@ cglobal h264_idct_add8_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride,
     call         h264_idct_add8_mmx_plane
     RET
 
+cglobal h264_idct_add8_422_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
+; dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
+    movsxdifnidn r3, r3d
+%ifdef PIC
+    lea     picregq, [scan8_mem]
+%endif
+%if ARCH_X86_64
+    mov       dst2q, r0
+%endif
+
+    mov          r5, 16  ; i
+    add          r2, 512 ; i * 16 * sizeof(dctcoef) ; #define dctcoef int16_t
+
+    call         h264_idct_add8_mmx_plane
+    add r5, 4
+    call         h264_idct_add8_mmx_plane
+
+%if ARCH_X86_64
+    add       dst2q, gprsize ; dest[1]
+%else
+    add        r0mp, gprsize
+%endif
+
+    add r5, 4   ; set to 32
+    add r2, 256 ; set to i * 16 * sizeof(dctcoef)
+
+    call         h264_idct_add8_mmx_plane
+    add r5, 4
+    call         h264_idct_add8_mmx_plane
+
+    RET
+
 h264_idct_add8_mmxext_plane:
     movsxdifnidn r3, r3d
 .nextblock:
diff --git a/libavcodec/x86/h264dsp_init.c b/libavcodec/x86/h264dsp_init.c
index 027c1ae0b3..ed52c4d8ef 100644
--- a/libavcodec/x86/h264dsp_init.c
+++ b/libavcodec/x86/h264dsp_init.c
@@ -78,6 +78,8 @@ IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
 
+IDCT_ADD_REP_FUNC2(, 8_422, 8, mmx)
+
 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
 
@@ -228,8 +230,11 @@ av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
 
             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
-            if (chroma_format_idc <= 1)
+            if (chroma_format_idc <= 1) {
                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
+            } else {
+                c->h264_idct_add8 = ff_h264_idct_add8_422_8_mmx;
+            }
             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
             if (cpu_flags & AV_CPU_FLAG_CMOV)
                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;